WardZero

People recovering at home still need someone to notice a falling oxygen level or a rising heart rate before it becomes an emergency. Families and a small care team cannot watch every reading. WardZero is a home-ward console that collects simulated sensor readings, explains the risk, and lets the care team acknowledge an alert and record what they did.

I built a React and Express app on PostgreSQL. A software sensor loop sends heart rate, SpO₂, temperature, and blood pressure. Risk is a readable rule over the recent trend, with an on-screen disclaimer that it does not diagnose. Alerts can be acknowledged. Caregiver notes are matched in Hindi and English. I did not claim a clinical accuracy number, and I did not require hardware.

The running app shows the path the brief asks for: sensor, platform, risk level with a reason, alert, and caregiver response. Admin, doctor, nurse, and caregiver each see only their patients. Setup is in README.md. There is no accuracy percentage because the risk rule is not a validated model.

I deployed WardZero as one public site on Render, with Neon as the PostgreSQL database. The website and the API share that URL so sign-in stays on one address. I did not use Vercel for the whole app, because the sensor loop has to keep running. Vercel sleeps between requests, so Live Monitoring would stop.

The first Render start failed. Neon was an empty database, so the users table did not exist yet. I changed startup so the core tables are created on an empty database, then seeded the demo accounts. A Postgres SSL warning on the hosted connection string was handled by using verify-full. After that, the live site opens in the browser and the demo roles can sign in. No hardware is required. The sensors are simulated.

I would add real email for password reset. Today the reset code is shown on screen because this ward has no mail server. Phone calls and SMS still depend on a Twilio trial, which can reach only a verified number. The risk level is a readable rule over the recent vital trend, not a trained or validated clinical model, so I am not claiming an accuracy percentage. On the free Render plan the first open after idle time can take about a minute.

I learned that a cloud database does not arrive with the tables a local Postgres already had, and that a login cookie only works when the site and the API are deployed together.